航空学报 > 2025, Vol. 46 Issue (2): 330696-330696   doi: 10.7527/S1000-6893.2024.30696

基于分解优化并行ESN的氢燃料电池寿命预测

华志广1, 潘诗媛1(), 赵冬冬1, 李祥隆2, 窦满峰1   

  1. 1.西北工业大学 自动化学院,西安 710129
    2.北京化工大学 材料科学与工程学院,北京 100029
  • 收稿日期:2024-05-17 修回日期:2024-06-11 接受日期:2024-07-24 出版日期:2024-08-21 发布日期:2024-08-20
  • 通讯作者: 潘诗媛 E-mail:panshiyuan@mail.nwpu.edu.cn
  • 基金资助:
    国家自然科学基金(52307251);中国博士后科学基金(2023TQ0277)

Lifespan prediction of hydrogen fuel cell based on decomposition optimization parallel ESN

Zhiguang HUA1, Shiyuan PAN1(), Dongdong ZHAO1, Xianglong LI2, Manfeng DOU1   

  1. 1.School of Automation,Northwestern Polytechnical University,Xi’an 710129,China
    2.School of Materials Science and Engineering,Beijing University of Chemical Technology,Beijing 100029,China
  • Received:2024-05-17 Revised:2024-06-11 Accepted:2024-07-24 Online:2024-08-21 Published:2024-08-20
  • Contact: Shiyuan PAN E-mail:panshiyuan@mail.nwpu.edu.cn
  • Supported by:
    National Natural Science Foundation of China(52307251);China Postdoctoral Science Foundation(2023TQ0277)

摘要:

针对质子交换膜燃料电池(PEMFC)多时间尺度老化特性导致电压预测精度较低的问题,基于集成经验模态分解(EEMD)与循环系统优化(CSBO)方法,提出了一种并行回声状态网络(PESN)结构,提升了PEMFC的寿命预测精度。采用EEMD对原始电压信号进行模态分解,将不同时刻的历史数据及分解得到的不同频率信号作为ESN不同子蓄水池的并行输入,构建一种按权重分配叠加输出的并行ESN结构,利用CSBO优化并行ESN结构的相关参数,基于优化后的EPESN模型实现PEMFC未来数百小时输出电压的预测。在稳态和准动态70%的数据训练集下,EPESN比ESN的均方根误差分别降低了34.25%和47.41%。在动态1训练时长为300 h时,EPESN比ESN的均方根误差降低了15.30%。结果表明:EPESN结构能够提高PEMFC寿命的预测精度。

关键词: 质子交换膜燃料电池, 寿命预测, 经验模态分解, 循环系统优化, 回声状态网络

Abstract:

Aiming at the problem of low voltage prediction accuracy caused by multi-time-scale aging characteristics of Proton Exchange Membrane Fuel Cell (PEMFC), a Parallel Echo State Network (PESN) structure which based on Ensemble Empirical Mode Decomposition (EEMD) and Circulatory System based Optimization (CSBO) is proposed to improve the lifespan prediction accuracy of PEMFC. By utilizing EEMD to conduct modal decomposition on the original voltage signal, the historical data from various time points and the decomposed signals with distinct frequencies are taken as the parallel inputs for the different sub-reservoirs of ESN, thereby establishing a parallel ESN structure that allocates and superimposes outputs based on weights. The CSBO is then leveraged to optimize the relevant parameters of the parallel ESN structure. Subsequently, utilizing the optimized EPESN model, the prediction of the output voltage of PEMFC for the next several hundred hours is achieved. Specifically, under the steady-state and quasi-dynamic data training sets at 70%, the Root Mean Square Error (RMSE) of EPESN is reduced by 34.25% and 47.41% respectively, compared to that of ESN. Furthermore, when the dynamic 1 training duration is set at 300 h, the RMSE of EPESN is decreased by 15.30% compared to ESN. The results explicitly demonstrate that the EPESN structure is capable of enhancing the prediction accuracy for the lifespan of PEMFC.

Key words: proton exchange membrane fuel cell, life prediction, empirical mode decomposition, circulatory system based optimization, echo state network

中图分类号: